It is 10:47 on a Tuesday. A woman named Camila — a premium banking customer for 11 years, with two mortgage products, a black card, and an active investment profile — is on the phone. It is her third call this week. The first was to the WhatsApp bot. The second was to the IVR. This one is to the supervisor.
For the fourth time in six days, someone on the other end asks her: “To validate your identity, could you please confirm your ID number and the last four digits of your card?”
Camila takes a breath. She provides the information. She explains again that the duplicate charge from the 19th of last month was never reversed, that she has already been told three times that someone would call her back, and that she needs to speak with someone who has the full case history. The supervisor takes notes, promises that someone will call her back tomorrow, and hangs up.
Camila does not hang up angry. She hangs up decided. That night, after putting her children to bed, she opens the app of a competing bank and begins the process of moving her payroll account.
The bank’s report that week will record three handled conversations with Camila. Average handling time: good. First contact resolution rate: acceptable. Post-call CSAT: she did not answer the survey. The executive committee dashboard on Monday will be green.
What that dashboard will not show is that the bank has just lost a customer whose net present value is several times higher than the entire annual investment of the Customer Experience department.
This is not fiction. It is what is happening right now, at this very moment, across hundreds of operations in Latin America. And the most celebrated metric in your CX department is exactly the one disguising the leakage.
“Handled Conversations”: The Metric That Rewards Fragmentation
Most CX, collections, and sales departments measure interactions by channel. Calls answered, chats resolved, bots activated. The problem is that customers do not live in silos. They live in one single flow. And every time they switch channels and have to start over, their experience pays the cost of an architecture that prioritizes touchpoints over continuity.
The data is brutal. According to Zendesk’s CX Trends report, 74% of customers find it frustrating to repeat their story to different agents. McKinsey quantifies it even more sharply: 56% of customers say they have to repeat themselves during support interactions. And Gartner estimates that companies lose an average of $12.9 million per year due to poor data quality and information silos.
In purely operational terms, high-effort interactions cost 37% more than low-effort ones. But the real impact is on loyalty: 96% of customers who experience high-effort interactions become disloyal, compared to only 9% in seamless experiences. Every repetition is an invisible failure that bleeds money and customers at the same time.
The “handled conversations” metric hides this bleeding. It makes the operation look productive while loyalty erodes silently.
The Problem Is Not Your AI. It Is That You Have Too Many AIs That Do Not Talk to Each Other.
Here is the uncomfortable part. Most large companies are not suffering from a lack of AI. They are suffering from an excess of poorly connected AI. One bot for service, another for collections, another for sales, an RPA layer in the back office, a copilot inside the CRM. All intelligent separately. None aware of what the other just did.
Gartner predicts that more than 40% of agentic AI projects will be canceled before 2027 due to uncontrolled costs, nonexistent ROI, and what Gartner calls “agent washing”: vendors rebranding chatbots and RPA as autonomous agents without real orchestration. Of the thousands of “agentic vendors” in the market, only about 130 are real, according to the firm.
The adoption of AI in silos did not solve operational fragmentation. It disguised it with new vocabulary. Connecting bots with duct tape is not architecture. It is technological theater. And Camila, that night in the competitor’s app, is the real cost that this theater produces.
The Real Metric: Operational Continuity
If “handled conversations” is the metric that lies, which one tells the truth?
Just one: how many times did the customer have to start over?
That number, measured honestly, is the true KPI of operational maturity. It means there is shared memory between humans and AI agents. It means that when a case escalates, the person on the other end does not open with “tell me again,” but with “I saw that you already spoke with the assistant about the duplicate charge; let me resolve it now.” It means every interaction is a continuation, not a restart.
That is what McKinsey calls the real leap of omnichannel: moving away from retaining data by channel and toward unifying the customer’s operational context. Companies that achieve this reduce service costs by 3–7%. But the real impact lies elsewhere: in an operation where every exception resolved by a human today becomes automated system logic tomorrow. A factory that optimizes itself.
Teams already operating under this philosophy do not buy “AI bots.” They buy resolution operating systems, where AI agents and humans solve cases as a single team, with shared memory and versioned governance. The difference is not semantic. It is architectural. And it is the only difference that would have kept Camila as a customer of the bank today.
The Real Inflection Point
Next quarter, another language model will arrive: more powerful, cheaper, and more accessible. All your competitors will have it too. The question is no longer which AI you choose. The question is how well you orchestrate the AI you already have with everything that already exists in your operation: your humans, your systems, your customer memory, and your governance policies.
Because as long as we keep rewarding volume over continuity, we will keep building operations that look intelligent in the report and feel broken on the call.
Camila hung up eight minutes ago. The next time your dashboard shows a record number of “handled conversations,” ask yourself: how many of those conversations were with a customer who had already decided to leave?
